Show simple item record

dc.contributor.authorCasas Manzanares, Noé
dc.contributor.authorRodríguez Fonollosa, José Adrián
dc.contributor.authorRuiz Costa-Jussà, Marta
dc.contributor.otherUniversitat Politècnica de Catalunya. Departament de Teoria del Senyal i Comunicacions
dc.date.accessioned2018-05-14T14:18:09Z
dc.date.available2018-05-14T14:18:09Z
dc.date.issued2018
dc.identifier.citationCasas, N., Fonollosa, José A. R., Ruiz, M. A differentiable BLEU loss. Analysis and first results. A: International Conference on Learning Representations. "ICLR 2018 Workshop Track: 6th International Conference on Learning Representations: Vancouver Convention Center, Vancouver, BC, Canada: April 30-May 3, 2018". 2018.
dc.identifier.urihttp://hdl.handle.net/2117/117201
dc.description.abstractIn natural language generation tasks, like neural machine translation and image captioning, there is usually a mismatch between the optimized loss and the de facto evaluation criterion, namely token-level maximum likelihood and corpus-level BLEU score. This article tries to reduce this gap by defining differentiable computations of the BLEU and GLEU scores. We test this approach on simple tasks, obtaining valuable lessons on its potential applications but also its pitfalls, mainly that these loss functions push each token in the hypothesis sequence toward the average of the tokens in the reference, resulting in a poor training signal.
dc.language.isoeng
dc.rightsAttribution-NonCommercial-NoDerivs 3.0 Spain
dc.rights.urihttp://creativecommons.org/licenses/by-nc-nd/3.0/es/
dc.subjectÀrees temàtiques de la UPC::Informàtica::Intel·ligència artificial::Llenguatge natural
dc.subject.lcshNatural language processing (Computer science)
dc.subject.otherDifferentiable
dc.subject.otherBLEU
dc.subject.otherGLEU
dc.subject.otherNMT
dc.subject.otherSeq2seq
dc.titleA differentiable BLEU loss. Analysis and first results
dc.typeConference report
dc.subject.lemacTractament del llenguatge natural (Informàtica)
dc.contributor.groupUniversitat Politècnica de Catalunya. VEU - Grup de Tractament de la Parla
dc.description.peerreviewedPeer Reviewed
dc.relation.publisherversionhttps://openreview.net/forum?id=HkG7hzyvf
dc.rights.accessOpen Access
local.identifier.drac22027776
dc.description.versionPostprint (published version)
dc.relation.projectidinfo:eu-repo/grantAgreement/MINECO/1PE/TEC2015-69266-P
dc.relation.projectidinfo:eu-repo/grantAgreement/AEI/PCIN-2017-079
local.citation.authorCasas, N.; Fonollosa, José A. R.; Ruiz, M.
local.citation.contributorInternational Conference on Learning Representations
local.citation.publicationNameICLR 2018 Workshop Track: 6th International Conference on Learning Representations: Vancouver Convention Center, Vancouver, BC, Canada: April 30-May 3, 2018


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record

Attribution-NonCommercial-NoDerivs 3.0 Spain
Except where otherwise noted, content on this work is licensed under a Creative Commons license : Attribution-NonCommercial-NoDerivs 3.0 Spain